On the empirical performance of an additive random number generator

• Autores: William M. Fleischman
• Localización: Mini and Microcomputers and their applications / Emilio Luque Fadón (ed. lit.), 1988, págs. 17-20
• Idioma: inglés
• Texto completo no disponible (Saber más ...)
• Resumen

  • We provide a summary of a fairly substantial series of statistical tests we have carried out to validate the usefulness of a simple random number generator of the additive congruential type. This type of random number generator is particularly attractive for microcomputer applications because integer random values are produced without recourse to multiplication. This paper is intended to document a series of a tests incluiding those for equidistribution (chi-square and Kolmogorov-Smirnov), serial tests, tests for gaps and runs, the coupon collecto´s and poker tests and serial correlation tests which we have applied to one such random number generation of normally-distributed random values which can be implemented to return tandom values from the standard normal without the use of multiplications or transcendental funtions calls.

    We have used the generator and the methods described herein extensivelu in simulation studies and in the generation of sample point sets for investigating tthe properties of geometric algorithms. They are very attractive for microcomputer implementation because, in addition to eliminating the de need for multiplications, they impose relatively low overhead for storage and access and because additive congruential generators with very long periods are readily constructed.